How does placing imputation and feature selection inside a searched Pipeline prevent leakage?
Classic MLLevel: IntermediateFold-local preprocessing
How does placing imputation and feature selection inside a searched Pipeline prevent leakage?
Hints
- Cross-validation clones the complete estimator.
- Selection can use target information.
Answer
Each CV training fold fits its own preprocessing and selection before scoring its held-out fold. Precomputing these steps on all rows lets validation data influence fitted transformations.
Source and licence: scikit-learn 1.7.2 — compose · BSD-3-Clause · answer: Quantoros editorial